We have heard a lot about artificial intelligence (AI) over recent years, and its potential on humanity and the planet. AI has the ability to change many aspects of human life, and it could lead to massive changes in various industries.

Remember, AI is not about replacing humans. Instead, it offers the opportunity for humans to delegate mundane or repetitive tasks and concentrate on tasks that require more creative, human skills.
Before we continue further, what exactly is AI?
AI is artificial intelligence. Using AI is changing the way we work, live and are entertained.
In our homes, we might use AI to –

• Select music
• Turn off and on our lights, TV, radio, heating or a host of other things
• Set reminders or keep a calendar of events
• Vacuum the house
• Send messages
  
  

AI is becoming increasingly important in many different industries. In the UK, for example, the government announced £100 million in funding to develop the use of AI in agriculture and other industries. The European Union (EU) is establishing legislation to regulate the development and use of AI across a broad spectrum of industries, particularly health, education and finance. This legislation focuses on the ethics around how data is selected, collected and used. 

AI is increasingly used to teach machines how to operate in different contexts. In a factory, AI can be used to make a machine increasingly more effective at doing it’s intended job. Vehicles can be trained using AI to perform autonomous operations, to the point of being driverless. 

A knowledge of AI is beneficial for many organisations and businesses. A clear understanding of the EU’s developing legislation may be essential knowledge for companies based in (or trading with) EU countries. Aspects of this legislation may be adopted elsewhere if it successfully addresses the concerns raised by the public.

Therefore, a knowledge of AI is also useful for professionals across a range of skills and industries. Understanding the technological advances in AI, and its potential uses, can help professionals to stay ahead of the crowd. A deep understanding of the implications of AI will also help them to prepare for the social, economic and cultural changes that could arise following AI’s integration into multiple workplaces.

Deep learning, machine learning and neural networks

The terms deep learning, machine learning and neural networks are sometimes used interchangeably, although there are slight variations in their meanings. All three terms relate to artificial intelligence, but they are subsets of each other. Neural networks are part of machine learning and deep learning is part of a neural network.

Machine learning uses a combination of data produced by humans and a set of algorithms to mimic how humans learn. This process allows the machine to improve the accuracy of its work as additional data is provided. The more data samples that the machine has access to, the better it will get at making predictions about what might happen (with more data) or what the audience might respond positively towards. Normal machine learning requires humans to select the features which will be used to classify any data addressed to answer a specific question. There are 3 aspects to a machine learning process:

• The decision – the machine uses input data to make a pattern. This pattern will allow the machine to either classify or predict something. 
• An error function – the machine compares the pattern it has made with any known examples. This comparison helps the machine to decide how accurate the model may be.
• An optimisation process – the machine compares the model produced to the training set of data. This process will continue across multiple sets of data until accuracy is established.
  
  

Machine learning is integral in speech-to-text applications and speech recognition software (such as Siri). Early versions of these applications required samples of speech to be effective. This was not always successful with non-standard accents, although the various apps have improved with time (and more data).

There are four kinds of machine learning models:

• Supervised learning – data is labelled in order to train the machine to become accurate at classifying or predicting tasks. Supervised learning is used as part of cross-validation to solve real problems. For example, supervised learning would be used to identify spam emails and divert them to a spam folder.
• Unsupervised learning – data is not labelled. This requires the machine to analyse and group the data produced. Unsupervised learning relies on the machine to explore the data and look for similarities that could inform the humans involved. For example, unsupervised learning might be involved in AI making recommendations about potential books to read based on previous purchases or similar buying patterns by other readers.
• Semi-supervised learning – this uses a smaller data set to classify or make predictions. For example, the machine may limit recommendations of potential books to only one genre, or limit the selection choices to only books currently on sale.
• Reinforcement learning – this model isn’t trained with sample data but relies on trial and error. Successful outcomes allow the machine to predict the most likely response.
  
  

Artificial neural networks contain input, hidden and output layers. They function similar to neurons in a human body. A neural node is stimulated when it receives information. When the information load is extensive, the node will pass the information on to the next layer.

Deep learning uses unstructured, or raw, data to determine which sets of features are relevant to the categories chosen for consideration. The term ‘deep’ is used to indicate how many layers exist in a neural network. The most important aspect of a deep learning structure is that the artificial neural networks allow the machine to learn on its own and to make informed decisions based on the data it receives.  Some of the most important types of deep learning involve:

• Convolutional neural networks – these are particularly good at processing data that involves images. For example, in facial recognition technology/
• Recurrent neural networks – these learn by repetition. For example, a GPS device will make recommendations about alternative routes during heavy traffic. The device may also recognise the drivers’ preferred routes and choose those routes as a starting point.
  
  Deep learning can be useful when handling a lot of complex data. It can help to establish patterns in material which might not be visible in normal data analysis methods. However, deep learning requires a large amount of computer memory plus additional processing units. This means that deep learning models can be expensive and time consuming to install. The accuracy of any deep learning will depend on the quality of the available data. This raises questions about the accuracy of the interpretations made and about the quality of the data sources. Data security is also an ongoing concern for AI programming.